Download failure leaves invalid file

I just ran into an issue where Github failed to serve up a file properly, and that prevented all future runs of my cookbook from converging successfully.

The github_asset resource downloads directly into the destination file (/var/chef/cache/github_assets/org/repo/version/foo-version.tar.gz). If the download fails it leaves a zero-length archive behind with that name, and future cookbook runs don't re-download the file because the zero-length file exists. See https://github.com/reset/github-cookbook/blob/master/libraries/github_asset.rb#L94.

Suggested solution: download to a temporary file and rename it if the download completes successfully. I suggest downloading to a temporary name in the same directory so that the rename step doesn't also require a copy, if (for example) the /tmp directory is on a different volume than the /var/chef/cache/github_assets directory.

Start by reading libraries/github_asset.rb around line 94, where the download writes to the destination path. Verify the failure behavior and the existing resource flow, then ensure a failed download does not leave the destination archive while a successful download produces it through the temporary-file flow described in the issue.